Role of the ocean observing system in an end-to-end seasonal forecasting system

Скачать 101.67 Kb.
НазваниеRole of the ocean observing system in an end-to-end seasonal forecasting system
Размер101.67 Kb.
  1   2   3   4   5   6

Role of the ocean observing system in an end-to-end seasonal forecasting system

Magdalena A. Balmaseda (1), Yosuke Fujii (2), Oscar Alves(3) , Tong Lee(4) , Michele Rienecker(5),Tony Rosati(6) , Detlef Stammer(7), Yan Xue(8), Howard Freeland(9), Michael J. McPhaden(10), Lisa Goddard(11) , Caio Coelho(12)

(1) ECMWF, Shinfield Park, Reading RG2 9AX (UK),

(2)MRI, 1-1 Nagamine, Tsukuba, Ibaraki, 305-0052 (Japan),

(3) CAWCR , GPO Box 1289, Melbourne, VIC 3001, Australia

(4) NASA /JPL,4800 Oak Grove Dr.Pasadena, CA 91109,(USA) ,

(5)GMAO, NASA/GSFC, Greenbelt, MD 20771 (USA),

(6) NOAA/GFDL 201 Forrestal Road, Princeton, NJ 08540-6649 (USA),

(7)KlimaCampus Universität Hamburg, Bundesstr. 53, 20146 Hamburg (Germany),

(8)NOAA/NCEP, 5200 Auth Rd, Camp Springs, MD 20746 (USA),

(9)FOC, Institute of Ocean Sciences, Sidney, BC,V8L 4B2, (Canada),

(10)NOAA/PMEL 7600 Sand Point Way NE Seattle, Washington 98115 (USA),

(11)IRI,Lamont Campus, 228 Monell Bldg. 61 Route 9W, Palisades, NY 10964-8000 Lisa.Goddard@iri.columbia,edu

(12)CPTEC/INPE, Rod. Presidente Dutra, Km 40, SP-RJ, Cachoeira Paulista, SP (Brazil),


There is clear demand for reliable forecasts of climate at seasonal time scales for a variety of societal applications. This paper discusses the role of ocean observations in the different components of a seasonal forecasting system, namely the initialization of the ocean, coupled model development and calibration of model output, concluding that the maintenance and enhancement of the current observing system is of paramount importance for further progress in seasonal forecasting.

It is shown that the assimilation of ocean observations improves the skill of seasonal forecasts. Results indicate that no observing system is redundant. Independent observations, not directly assimilated, are necessary for the improvement of assimilation methods and numerical models, thus increasing the information content of the observations. Forecast calibration requires long observational records to produce historical ocean initial conditions. These are equivalent to ocean re-analyses, which, continuously brought up to real-time, allow the monitoring of relevant climate variables.

The current forecasting systems are not making optimal use of the existing observations, in particular in regions where model error is large and/or where the initialization is inadequate. This is particularly noticeable in the equatorial Atlantic. Improvements in numerical models and initialization strategies are needed to exploit the full potential of current and future observing systems.


Good-quality seasonal forecasts with reliable uncertainty estimates are of great value to society, allowing institutions and governments to plan actions to minimize risks, manage resources and increase prosperity and security. Human and economic losses that may be caused by adverse climate events can be mitigated with early warning systems (e.g. famine, epidemics) and disaster preparedness. Equally, adequate planning can aid the exploitation of favourable climate conditions.

Seasonal forecasts predict variations in the atmospheric circulation in response to anomalous boundary forcing [1], such as that provided by variations of sea surface temperature (SST) and land conditions (snow depth, soil moisture). Of special importance are the variations of the tropical SST, which have the potential to alter the large-scale patterns of atmospheric circulation associated with tropical convective cells. Thus, the predictability of climate variability on seasonal time-scales relies largely on the initial conditions of the model ocean.

Seasonal forecasting is currently a routine activity in several operational centres, with a growing number of economic and societal applications especially in the agriculture, health and energy sectors. The consolidation of seasonal forecasting over the last decade has been possible thanks to the improvement in coupled models and data assimilation methods, availability of atmospheric fluxes from reanalyses, and the development of the ocean observing system. In particular, the implementation of the full TAO/TRITON array in the Equatorial Pacific during the 10-yr (1985-94) Tropical Ocean Global Atmosphere (TOGA) program has been instrumental in advancing prediction of the El Niño/Southern Oscillation (ENSO), which is still considered the corner-stone of seasonal forecasting. The skill of seasonal forecasts has further improved with the advent of satellite altimeters and Argo. There is potential for improving the prediction of other modes of inter-annual variability, such as the Indian Ocean Dipole [2], which will benefit from the on-going development of the Indian Ocean observing system. The potential of the current observing system for seasonal forecasts has not yet been fully exploited, and further progress is expected. It is essential that we maintain the current observing system in the years to come.

This paper discusses the value of the ocean observing system in an end-to-end seasonal forecasting system. Section 2 offers a brief overview of the importance of ocean variability for the prediction of the local and regional climate variability that impacts society. The current understanding of the role of ocean observations in developing and implementing end-to-end seasonal forecasting systems is discussed in section 3. Section 4 provides a brief discussion of how the information from dynamical forecast systems can assist decision makers. However, more work is still needed in improving and completing the observational systems, in improving the assimilation methods that ingest the observations and in improving the models that seek to capture the relevant processes. Thus, the paper concludes with an outlook for the next decade, offering a perspective on the major challenges ahead and a set of recommendations for future developments of the ocean observing system and its use from a seasonal forecasting perspective.


The dominant climate fluctuations at interannual time scales are related to ENSO, a quasi-periodic warming of sea surface temperatures in the eastern and central equatorial Pacific affecting the patterns of temperature and rainfall in much of the world [3]. ENSO plays a dominant role in the climate anomalies over the land areas surrounding the entire Pacific basin. The effects of ENSO are also noticeable in other tropical and extra-tropical regions via the so-called atmospheric bridge [4],[5], in, for example the Indian monsoon, Atlantic hurricanes and the climate of southern and eastern Africa. The importance of ENSO in seasonal forecasts is further enhanced by its potential predictability [6]. It has been shown that the most predictable variations in worldwide precipitation at interannual time-scales are related to ENSO [7].

Anomalies in SST other than ENSO can also drive temperature and precipitation anomalies on seasonal time-scales. Examples include the connection of the tropical Atlantic with north-east Brazil rainfall [8], [9] and the rainfall in west Africa and Sahel [10], [11], the impact of the extratropical Atlantic (e.g. [12]) on European climate, and the tropical Indian Ocean [13] (in particular the mode of variability known as the Indian Ocean Dipole [3], [14]) impact on east African rainfall and the Indian monsoon. The warming of SST in the tropical Indian after El Niño enhances the anticyclone in the Philippine Sea and impacts the climate in the east Asia [15]. Notable impacts of Pacific and Indian Ocean SST on the US droughts have been reported [16].

Apart from SST, there are other sources of seasonal predictability. The memory provided by snow depth and soil moisture should also be considered in seasonal forecasting systems. Studies have shown that increased concentration of greenhouse gasses also has a signature on seasonal forecasts [17]. More recent studies point to the role of the stratosphere in increasing seasonal predictability [18].


Seasonal forecasting systems are based on coupled ocean-atmosphere general circulation models that predict both the SSTs and their impact on the atmospheric circulation. Seasonal forecasting is considered an initial-value problem, in that the information provided by the initial conditions (especially the ocean) determines the predictability of the system. The chaotic nature of the atmospheric response is taken into account by issuing probabilistic forecasts, obtained by performing an ensemble of coupled integrations. Because of deficiencies in coupled models, the forecasts need calibration before the forecast is issued. A calibration is done by conducting a series of retrospective seasonal hindcasts, which in turn requires ocean initial conditions for a historical period (typically 15-25 years), equivalent to an ocean reanalysis. The hindcasts are also needed for skill assessment.

The generation of ocean initial conditions is the first step in a seasonal prediction system. Assimilation of observations into an ocean model forced by prescribed atmospheric fluxes is the most common practice for initialization of the ocean component of a coupled model. The emphasis is on the initialization of the upper ocean thermal structure, particularly in the tropics, where SST anomalies have a strong influence on the atmospheric circulation.

The information from the initial conditions is projected into the future by forward integration of numerical ocean-atmosphere general circulation models. To sample the inherent uncertainty of seasonal predictions, model integrations include an ensemble of forecasts from slightly perturbed initial conditions or model formulations. The quality of the coupled model is critical for achieving accurate seasonal forecasts. Observations of the ocean and atmosphere have contributed to the understanding and parameterization of relevant processes, leading to the improvement of coupled models. For instance, Fig. 1 shows that the improvements in ENSO forecasts at the ECMWF over the past decade. The improvements can be attributed equally to better initialization of the ocean and improved coupled models.

Figure 1. Progress in the seasonal forecast skill of the ECMWF operational system during the last decade. The solid bar shows the relative reduction in mean absolute error of forecast of SST in the eastern Pacific (NINO3). The brown-striped bar shows the contribution from the ocean initialization, and the white-striped bar is the contribution from model improvement.

In spite of the improvements, forecasts from a single forecasting system are often not reliable enough. This is especially true for seasonal forecasts of precipitation: deficiencies in model formulation result in overconfident forecasts, in the sense that the ensemble spread often does not include the verifying observation. Ensemble generation techniques that sample model uncertainty (multi-model ensemble), and that are efficient at capturing the coupled model growing modes (i.e. breeding vectors) are needed. In addition, a posteriori calibration procedures are used in an attempt to obtain reliable forecast products.

The following sections discuss how improvements in end-to-end dynamical seasonal forecasting systems rely on three interconnected efforts: (1) assimilation and initialization methods, (2) process studies and model improvements, and (3) assessment and verification.

3.1 Initialization of the ocean and ocean re-analyses

The simplest way to initialize the ocean is to run an ocean model forced with winds and fresh water fluxes and with a strong relaxation of the model SST to observations. This technique would be satisfactory if errors in the forcing fields and ocean model were small. However, surface flux products and ocean models are both known to have significant errors. The uncertainty induced in the upper ocean by using different wind products can be as large as the interannual variability. Assimilation of ocean observations is then used to constrain the estimation of the ocean state.

Sea surface temperature observations are essential. Most of the initialization systems also use subsurface temperature (from XBT’s [19], moored buoys [20] and Argo [21]), most recently also salinity (mainly from Argo), and altimeter-derived sea-level anomalies (SLAs) [22]. The latter usually need a prescribed external Mean Dynamic Topography (MDT), which can be derived indirectly from gravity missions such as GRACE and, in the near future, GOCE [23]. Some of the initialization systems use an on-line bias correction scheme or relaxation to climatology to control the mean state. An overview of ocean re-analyses (ORAs) systems used for initialization of operational or quasi-operational seasonal forecast systems is provided in [24].

The ocean re-analyses used for the initialization of seasonal forecasts are a valuable resource for climate variability studies and have the advantage of being maintained in near real-time, so that the time variability of relevant climate variables can be monitored. This complements the ENSO monitoring based on TAO/TRITON at Most of the operational systems offer real-time information about selected ocean fields and observation coverage. Fig. 2 shows time series of a proxy for upper ocean heat content anomalies (averaged temperature anomalies in the upper 300m) in selected areas from 1985 to present from seven different ORAs [25]. The anomalies were based upon the 1985-2002 climatology and smoothed with a 12-month running mean. The dispersion between the curves can be taken as a measure of uncertainty in our knowledge of the climate. The uncertainty in some indices is larger than others: the interannual variability of the Indian Ocean dipole and the decadal variability of the north nubtropical Atlantic seem to be robust among ORAs. All the ORAs also show warming trends in the North Atlantic and global ocean, but there is uncertainty in the magnitude of the trend. This uncertainty is large in later years and it is important to determine the origin of this uncertainty (observations used, data assimilation methods, models, forcing fields, etc.). Reducing uncertainty in the estimation of climate indices should be a high priority for the community. More sophisticated monitoring tools have been developed by the Climate Prediction Center (CPC) of the National Centers for Environmental Prediction (NCEP) in the USA to monitor and assess the Indian Ocean Dipole [3], tropical Atlantic variability [33] and Pacific Decadal Oscillation [34] (

Figure 2: Time series of averaged temperature anomalies in the upper 300m in selected areas from 1985 to present. Seven real-time operational ocean re-analyses are shown. From [25]

Major progress has been achieved during the past decade in the field of ocean data assimilation, largely stimulated by international coordination through GODAE [35]. The first generation of ocean initialization systems were univariate and assimilated only temperature data: the observations of temperature were used only to correct the model temperature field, leaving the other model variables untouched. These systems were able to reduce the uncertainty in the thermal structure, and sometimes would improve the forecast skill. However the resultant velocity and salinity fields were often degraded since the univariate assimilation procedure introduced dynamical inconsistencies. Nowadays most of the ocean initialization systems are second generation: they assimilate temperature, salinity and sea level via

NINO3 5ºS-5ºN, 90-150ºW

NINO34 5ºS-5ºN, 170-120ºW

NINO4 5ºS-5ºN, 160ºE-150ºW

EQ3 5ºS-5ºN, 150ºE-170ºW

EQPAC 5ºS-5ºN, 130ºE-80ºW

EQIND 5ºS-5ºN, 40º-120ºE

WTIO 10ºS-10ºN, 50º-70ºW

STIO 10ºS-0ºN, 90º-110ºE

EQATL 5ºS-5ºN, 70ºW-30ºE

NSTRATL 5ºN-28ºN, 80ºW-20ºE

NATL 30ºN-70ºN, 70ºW-15ºE

NPAC 30ºN-70ºN, 100ºE-100ºW

Table 1: Definition of area average indices

multivariate schemes, imposing physical and dynamical constraints among different variables. Results from several of these “second generation initialization systems” show that the assimilation of ocean data in the ocean initialization improves seasonal forecast skill, although ultimately, the impact of initialization in a seasonal forecasting system will depend on the quality of the coupled model [24], [36].

The skill of seasonal forecasts is often used to gauge the quality of the ocean initial conditions. This may not always be appropriate, since the quality of the coupled model is also important - if the major source of forecast error comes from the coupled model, improvements in ocean initial conditions would have little impact on forecast skill. This is something to bear in mind when interpreting results of the impact of ocean data assimilation on seasonal forecasts.

Several studies have demonstrated the benefit of assimilating ocean data on the prediction of ENSO [37], [38], [39], among others). The benefits are less clear in other areas, such as the equatorial Atlantic, where model errors are large and there is no long history of moored observations, as in the Pacific.

The initialization strategy can influence the mean and variability of seasonal forecasts. Ref [40], using the latest version of the ECMWF seasonal forecasting system (S3), evaluates three different initialization strategies, each of which uses different observational information. Strategy i) uses ocean, atmospheric and SST information, strategy ii) uses atmospheric information and SST, and strategy iii) uses only SST, as in [41]. In method (i), the coupled system thus starts close to the observed state but it is not obvious that this leads to the most skilful forecasts as the method can have undesirable initialization shocks. Method (iii) can reduce the initialization shock since the atmospheric and ocean models will be in closer balance at the start of the coupled integrations. The three experiments can also be seen as observing system experiments. Differences between (i) and (ii) are indicative of the impact of ocean observations, and comparison of (ii) and (iii) are indicative of the impact of the atmospheric observations that were used to produce the atmospheric reanalyses. Results show that the initialization strategy has an impact on both the mean state and the interannual variability of coupled forecasts. They also show that, in this particular system, initialization shock does not preclude forecast skill, and the most skilful forecasts are those obtained when the initial conditions are closer to the “real ocean state”, even if this causes sizable adjustment processes.

Fig. 3a shows the relative reduction in the monthly mean absolute error (MAE) resulting from adding information from the ocean and/or atmospheric observations for the 1-7 month forecast range in the regions defined in Table 1. Observational information has the largest impact in the western Pacific (EQ3), where the combined information of ocean and atmospheric observations can reduce the MAE more than 25% (50% in the first 3 months, not shown). With the exception of the equatorial Atlantic (EQATL), the best scores are achieved by strategy i). This means that for the ECMWF system, the benefits of ocean data assimilation and the use of fluxes from atmospheric (re)analyses more than offset problems arising from initialization shock.

Seasonal forecast skill can also be used to evaluate the ocean observing system. Fig. 3b shows the relative reduction in the 1-7 month forecast error by including information from the moored arrays, altimeter-derived sea-level anomalies and the mean dynamic topography (MDT) used as reference for the altimeter-derived anomalies. The statistics are for the period 1993-2006. The information from the mooring array is the dominant factor in improving skill in different regions of the equatorial Pacific and improves the skill in the equatorial Indian Ocean (likely a remote effect).The impact of the external MDT is also quite substantial in the Pacific, and to a lesser degree the equatorial Indian Ocean. The effect of altimeter data is more noticeable in the NINO3 and NSTRATL. Moorings, MDT and altimeters also have a positive impact on the WTIO, although the individual contributions are small. The equatorial Atlantic again stands out as the only region where the different observational information consistently has a detrimental effect, indicative of problems with either the assimilation system and/or the coupled model.

Fig. 3c shows the impact on forecast skill of Argo, moorings and altimeters. The statistics have been calculated only for the (rather short) Argo period 2001-2006 and so the impacts are best considered as indicative rather than definitive. The figure shows that no observing system is redundant. Argo has a dominant impact in the western Pacific (NINO4) and equatorial Indian Ocean. Argo is the only observing system with a significant positive impact on the WTIO and SETIO regions. The information from the moorings is still dominant in most of the equatorial Pacific, although in the NINO4 region it is less important than that from Argo. Meanwhile altimetry has a significant positive impact in the equatorial Pacific, and is the only observing system with positive impact in the north subtropical Atlantic. Again, for this period, all the observing systems have a negative impact on the EQATL region.

Figure 3: Impact of initialization in forecast skill for different regions, as measured by the reduction in mean absolute error for the forecast range 1-7 months. The different areas in the x-axis are defined in Table 1.

(a) Comparison of initialization strategies for the period 1987-2006. OCOBS indicates the impact of ocean observations. ATOBS indicates the impact of atmospheric observation, while OC + AT represents the combined impact of atmospheric and oceanic data. (b) Comparison of altimeter, moorings and MDT for the period 1993-2006. ALTI indicates the difference in skill between NO-ALTI and ALL, and MOOR the difference between NO-MOOR and ALL. MEAN indicates the differences from using the different MDTs. (c) Comparison between Argo, altimeter and moorings for the period 2001-2006. ARGO represents the difference between NO-ARGO and ALL. Only differences exceeding the 70% significant level of a one-tailed T-test are shown. From [40]

Figure 4: Depth-Time sections of salinity for the period 1999-2006 at 156E and 5N (upper row), Equator (central row) and 5S (lower row). The left column is for experiment NOS, where only temperature is assimilated and a balanced T-S relationship is not imposed. The second column, for experiment TH, is for the experiment when only temperature data is assimilated including the T-S relationship. The third column is for experiment ALL, where salinity and temperature are assimilated. The right column shows the observational value from the TRITON array. Vertical grid lines mark the beginning of each year. The horizontal grid line interval is 30 m. From [56].

The impact of the TAO/TRITON array and Argo float data has also been evaluated with the JMA seasonal forecasting system [42] by conducting data retention experiments for the period 2004-2007. The results (not shown) are consistent with the above ones, indicating that TAO/TRITON data improves the forecast of SST in the eastern equatorial Pacific (NINO3, NINO4), and that Argo floats are essential observations for SST prediction in the tropical Pacific and Indian Oceans.

  1   2   3   4   5   6


Role of the ocean observing system in an end-to-end seasonal forecasting system iconIf you read all the way through this issue of The Latest Stuff from Gerry there is a priceless reward awaiting you at the end. If you jump to the end your

Role of the ocean observing system in an end-to-end seasonal forecasting system icon16 (3), 107. Adams, V m and Plosker, g r (1997) Concepts in end-user training: how to convert end users into effective searchers, Searcher, 5

Role of the ocean observing system in an end-to-end seasonal forecasting system iconObserving System Design and Assessment

Role of the ocean observing system in an end-to-end seasonal forecasting system iconAnswer the following questions in writing and submit by end of class today. Finish the two page paper (see# 25) by end of class tomorrow

Role of the ocean observing system in an end-to-end seasonal forecasting system icon2. System Parameters for load forecasting. 6

Role of the ocean observing system in an end-to-end seasonal forecasting system iconBridging the Gap between Weather and Seasonal Forecasting: Intraseasonal Forecasting for Australia

Role of the ocean observing system in an end-to-end seasonal forecasting system iconA global Ocean Carbon Observation System

Role of the ocean observing system in an end-to-end seasonal forecasting system iconThe Place and Role of Official Statistics within the National System of Health Monitoring

Role of the ocean observing system in an end-to-end seasonal forecasting system iconJoint wmo technical progress report on the global data processing and forecasting system and numerical weather prediction research activities for 2008

Role of the ocean observing system in an end-to-end seasonal forecasting system iconJoint wmo technical progress report on the global data processing and forecasting system and numerical weather prediction research activities for 2011

Разместите кнопку на своём сайте:

База данных защищена авторским правом © 2014
обратиться к администрации
Главная страница